Bayesian inference of frequency-specific functional connectivity in MEG imaging using a spectral graph model.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Graph Theory

Background:

• Understanding the brain's structural connectivity (SC) and functional connectivity (FC) relationship is crucial.
• Existing complex neural mass models lack direct interpretation of structure-function relationships and require intensive computation.
• There is a need for intuitive and computationally efficient models to bridge SC and FC.

Purpose of the Study:

• To develop a novel method for mapping brain structural connectivity to functional connectivity using spectral graph theory.
• To create a parsimonious model with biophysical underpinnings that predicts functional connectivity.
• To validate the model's performance against existing methods and real-world neuroimaging data.

Main Methods:

• Employed spectral graph theory and graph harmonics (eigenvectors of the Laplacian matrix) to model brain activity.
• Developed a parsimonious spectral graph model (SGM) with only three global, spatially invariant parameters.
• Utilized a deep neural network-based simulation-based inference for efficient model parameter estimation from resting-state MEG data.

Main Results:

• The SGM successfully predicted functional connectivity (FC) in arbitrary frequency bands using graph harmonics.
• The model generated rich FC patterns with high accuracy using only a few harmonics.
• The proposed method demonstrated superior performance compared to benchmark methods like graph diffusion and coupled neural mass models.

Conclusions:

• Spectral graph modeling provides a powerful and intuitive framework for understanding the SC-FC relationship.
• A single biophysical parameterization can simultaneously fit functional connectivities across multiple frequency bands.
• This approach offers a computationally efficient and accurate tool for brain connectivity research.